Skip to content
    Case studiesChapter 89

    How McKinsey Plans to Survive AI (and Reinvent Consulting)

    Bob Sternfels on consulting's transformation: a human-plus-agent operating model, the shift to outcomes, and organizational design as the source of AI value.

    Film or talkadvanced30 minvolatile · reviewed Aug 13, 2026
    Engineering Manager
    Staff+
    Chapter outline

    Brief

    The essential idea

    In an approximately 31-minute HBR IdeaCast interview, McKinsey managing partner Bob Sternfels argues that AI rarely creates enterprise value by itself. The larger gain comes from redesigning end-to-end processes, removing handoffs, and establishing a human-plus-agent operating model with transparent result metrics.

    An agent becomes an operational asset with an owner, tasks, access boundaries, tools, and verifiable output. As analysis becomes commoditized, consulting and internal technology teams must shift from billing or celebrating activity toward measurable outcomes such as time to value, MTTR, cost to serve, quality, and business change.

    Speed remains useful only when paired with recoverability. Leaders need quality gates, human judgment, governance, auditability, and a registry of agents, while acknowledging open questions about regulated organizations, comparable agent metrics, and the development of junior engineers when AI absorbs entry-level work.

    Decision lens

    Key takeaways

    AI layered over an unchanged process usually automates steps without improving the whole system.

    The human-plus-agent pair is emerging as a new unit of productivity.

    Agents need owners, access rules, quality gates, and auditable results.

    Value shifts from volume of artifacts to measurable business outcomes.

    Faster organizations also need rapid detection, rollback, and recovery.

    Regulation and organizational constraints limit how far flattening and automation can go.

    Leaders must preserve opportunities for junior engineers to develop judgment.

    Workplace experiment

    Apply it at work

    1. 1

      Choose one or two end-to-end processes and redesign them around a human-plus-agent workflow rather than automating one step.

    2. 2

      Create an agent registry covering owner, task, data access, tools, quality gates, and audit trail.

    3. 3

      Replace activity measures with time-to-value, MTTR, cost-to-serve, quality, or another defined outcome.

    4. 4

      Use an Agent-to-PR-to-quality-gates-to-human-review pipeline for production changes.

    5. 5

      Design an explicit learning path for junior engineers whose former practice tasks are increasingly automated.

    Choose one action, define the observable effect, and keep the first test small enough to reverse.

    Evidence

    Sources and further reading

    Additional sources

    Channel, aggregator, and commentary links confirm the work; they are not the primary source.

    Previous chapterOpenAI's Head of Platform Engineering on the Next 12-24 Months of AINext chapterByteDance: The App Factory and Volcano Engine (Case Study)